6
Recommendations

Many advances in computer science and engineering in the last 10 to 20 years speak to the problems in health care information technology (IT) observed by the committee. These advances include ontologies, data fusion techniques, large-scale search capabilities, information visualization, and modern computer system architectures to support large-scale distributed systems in a heterogeneous operating environment. But for various reasons, these advances have not often been reflected in generally available clinical information systems.

Organizations face difficult economic decisions regarding whether to emphasize short-term financial gains relative to longer-term advantages wherein cost savings are associated with quality improvement. In addition, the acquisition processes of many health care provider organizations are not often compatible with the development and deployment of future health care IT systems that provide cognitive support and are evolvable into the future. Poorly understood or defined requirements, poor development processes, and failures to adopt iterative or evolutionary approaches or user-centered design are often seen.

In addition, it is fair to say that the integration of health care IT into operational work processes has proven both more essential and more difficult than was first expected, at least in part because many attempts to deploy health care IT have not taken into account the systems engineering issues inherent in viewing health care as a complex, adaptive system. In other words, the research problems have become significantly more demanding when conceptualizing the whole as a set of components



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 59
6 Recommendations Many advances in computer science and engineering in the last 10 to 20 years speak to the problems in health care information technology (IT) observed by the committee. These advances include ontologies, data fusion techniques, large-scale search capabilities, information visualiza- tion, and modern computer system architectures to support large-scale distributed systems in a heterogeneous operating environment. But for various reasons, these advances have not often been reflected in generally available clinical information systems. Organizations face difficult economic decisions regarding whether to emphasize short-term financial gains relative to longer-term advantages wherein cost savings are associated with quality improvement. In addi- tion, the acquisition processes of many health care provider organizations are not often compatible with the development and deployment of future health care IT systems that provide cognitive support and are evolvable into the future. Poorly understood or defined requirements, poor develop- ment processes, and failures to adopt iterative or evolutionary approaches or user-centered design are often seen. In addition, it is fair to say that the integration of health care IT into operational work processes has proven both more essential and more difficult than was first expected, at least in part because many attempts to deploy health care IT have not taken into account the systems engi- neering issues inherent in viewing health care as a complex, adaptive system. In other words, the research problems have become significantly more demanding when conceptualizing the whole as a set of components 

OCR for page 59
0 COMPUTATIONAL TECHNOLOGY FOR EFFECTIVE HEALTH CARE working together to provide a working information and knowledge infra- structure for 21st century health care. Lastly, there are many unsolved problems related to health care IT, including supporting appropriate access while respecting the confiden- tiality of medical records, managing the cognitive load on care provid- ers that results from the availability of large volumes of information, and managing the information in a medical record over the multidecade lifetime of individuals in the context of rapidly changing scientific and medical knowledge. Three distinct groups have a meaningful role in addressing these areas. Federal and state government and the health care community must speak to acquisition policy. The health care community must insist that vendors supply health care IT systems that provide meaningful cognitive support. And the research community, including researchers in computer science and health/biomedical informatics, must play a lead intellectual role in advancing the current state of the art in health care IT systems. 6.1 GOVERNMENT Federal and state governments play important roles as supporters of research, payers for health care, and stimulators for education. The com- mittee believes that government organizations—especially the federal government—should explicitly embrace measurable health care qual- ity improvement as the driving rationale for its health care IT adoption efforts, and should shun programs that focus on promoting the adoption of specific clinical applications. While this principle should not be taken to discourage incentives to invest in infrastructure (networks, worksta- tions, administrative transaction processing systems, platforms for data mining, data repositories, and so on) that provides a foundation on which other specific clinical applications can be built, a top-down focus on spe- cific clinical applications is likely to result in a premature “freezing” of inefficient workflows and processes and to impede iterative change. In focusing on the goal to be achieved, namely better and/or less expensive health care, clinicians and other providers will be eager to use new health care IT-enabled clinical applications if, where, and when such applications can be shown to enable them to do their jobs more effectively. Health care quality improvement efforts scale from practice groups and individual practitioners to large health care organizations to the health care system as a whole. Traditionally, quality improvement efforts tend to occur at the level of larger practice groups and health care organi- zations, and are slowed by the requirement to develop consensus among the universe of relevant clinicians. Indeed, these efforts require such vol- ume of collective effort that most organizations cannot sustain more than

OCR for page 59
 RECOMMENDATIONS a few quality improvement initiatives at a time. Given the quality chasm facing many health care organizations, such a slow rate of change is unacceptable. In contrast, iterative local improvement at the small group or even individual practitioner level has the major advantage of being faster and cheaper to accomplish because of its small scale. This allows for improve- ment efforts to be conducted in parallel, increasing the chances of finding successful approaches, while unsuccessful approaches can be rapidly and inexpensively discarded. Local successes also tend to build support for additional improvement efforts. Government should promote explora- tion of methods and models for small-scale improvement efforts as well as efforts to integrate these small-scale improvements on a larger scale. A balance with many small-scale efforts providing the evidence base for a smaller number of large-scale efforts seems appropriate. IT is a fundamental enabler for both large-scale and small-scale improvement efforts. But for the most part, the health care IT available in today’s market is not well suited to support small-scale optimization, which requires applications that are rapidly customizable in the field by end users. Federally inspired or supported initiatives that incentivize health care organizations to undertake iterative small-scale optimization, and subsequent translation of successes to a larger scale, are likely to help stimulate the creation of a new market for these applications—for example, such incentives might take the form of payment premiums for demonstrations of major improvement of a result (process or clinical) for a unit of the organization. A last point is that work at the health care–IT nexus is interdisciplin- ary. A lack of familiarity with the domain-specific problems in the health care domain has often impeded the efforts of well-meaning computer scientists. Formal and elegant computer science, as understood by most computer science researchers, is often a poor match with the complex cultural and organizational environment of health care and biomedi- cine—topics about which a well-trained computer science graduate is generally ignorant. Academic medical centers often fail to take advantage of relevant expertise—especially in health/biomedical informatics—that is available to them. Such organizations are often inclined to turn to inter- nal expertise—the in-house health care IT professionals—rather than to the relevant health/biomedical informatics and computer science faculty on campus. Progress at this nexus will require contributions of health care experts, computer science experts, experts from the health/biomedi- cal informatics community, and health care IT experts working together to understand the problems related to improving health care and how IT might be applied to address those problems.

OCR for page 59
 COMPUTATIONAL TECHNOLOGY FOR EFFECTIVE HEALTH CARE This analysis leads to six important recommendations for the federal government: • Incentiize clinical performance gains rather than acquisition of IT per se. This is not to say that IT is irrelevant, but the acquisition of health care IT is better guided by what is needed to support improvement efforts. 1 For example, the development and redesign of work processes to provide effective feedback to clinicians logically precede implementation of IT to automate workflow, rather than simply acquiring health care IT first. • Encourage initiaties to empower iteratie process improement and small-scale optimization. Because the market does not today provide the IT required for small-scale optimization (the committee saw no such health care IT in its site visits), these initiatives should also provide support for clinicians to work with computer science and IT experts to design pro- totype applications to support their improvement efforts. In this short report, the committee did not address the nature or scale of support needed, and believes that this is an issue best addressed in a second phase of this study. • Encourage deelopment of standards and measures of health care IT performance related to cognitie support for health professionals and patients, adaptability to support iteratie process improement, and effectie use to improe quality. One lever is to shift the focus of certification efforts from task-specific transactional capabilities to capabilities that provide better cognitive support for health professionals and patients. An example of a standard oriented toward cognitive support would be a requirement to test system effectiveness or human comprehension in the context of the data received by the system or person, perhaps in a simulation environ- ment or in an actual work environment. 1The federal government has two primary policy levers for promoting an agenda to im- prove health care quality—public reporting of comparative performance information and pay-for-performance payment policies. Both of these levers depend on the ability to aggre- gate and analyze data over entire patient episodes, and thus the federal government should require or incentivize submitting the data rather than specifying the particular health care IT to obtain it. Once the data are submitted, their aggregation and analysis can be accomplished through the kinds of health care IT described in Section 5.2.3. 2Standards are not a new idea in health care IT—indeed, they are a critical element of “plug-and-play” architectures that enable the infusion of new technologies when they are available (in contrast to monolithic architectures that make it difficult to take advantage of new technologies). However, to the best of the committee’s knowledge, standards oriented toward cognitive support essentially do not exist.

OCR for page 59
 RECOMMENDATIONS • Encourage interdisciplinary research in three critical areas:3 (a) orga- nizational systems-level research into the design of health care systems, processes, and workflow (i.e., research in systems engineering for a health care delivery context); (b) computable knowledge structures and models for medicine needed to make sense of available patient data including preferences, health behaviors, and so on; and (c) human-computer inter- action in a clinical context. a. Examples of process and workflow research include languages and systems to describe and visualize health care workflows; mod- eling of health care workflow at scale while enabling explicit step- by-step escalation rules; support for distributed, multiplayer deci- sion making among players with sometimes conflicting views of what factors are important; rigorous analysis and documentation of the workflow demands of routine practice to understand how computer technology could be used to facilitate and support the workflow of the practitioner; and use of queuing theory to opti- mize organizational performance. b. Examples of research into computable knowledge structures and models include computable guidelines and approaches for comparing, assessing, updating, and integrating these guidelines into a library of guidelines for a given patient; and systems that can infer clinical conditions from raw data (e.g., inferring that “patient is feeling more pain” from the report of an upward adjust- ment in the intravenous drip of a pain management drug). Because the clinical interpretation of data depends on the current state of knowledge about medicine and about physiology and how people respond to treatments and so on, computable structures are impor- tant because they connect medical knowledge to patient data in machine-readable and machine-executable form. Thus, they can 3It is beyond the scope of this report to describe in detail the infrastructure needed to sustain computer science research as it might apply to health care. However, the recommen- dations from another National Research Council report on research at the interface between computing and biology are instructive in this regard. That report indicated that . . . agencies and foundations should support awards that can be used for retraining purposes; balance quality and excellence against openness to new ideas in the review process; encourage team formation; provide research opportunities for investigators at the interface who are not established enough to obtain funding on the strength of their track record alone; use fund- ing leverage to promote institutional change; use publication venues to promote institutional change; support cyberinfrastructure for biological research; recognize quality publicly; recognize the costs of providing access to computing and information resources; define specific challenge problems that stretch the existing state of the art but are nevertheless amenable to progress in a reasonable time frame; work with other agencies; and provide the funding necessary to capital- ize on the intellectual potential of 21st century biology. (p. 383) See Chapter 11, National Research Council, Catalyzing Inquiry at the Interface of Computing and Biology, The National Academies Press, Washington, D.C., 2005.

OCR for page 59
 COMPUTATIONAL TECHNOLOGY FOR EFFECTIVE HEALTH CARE provide needed abstractions for the health care provider and the clinician to help them understand what is going on with a given patient. c. Examples of research into clinically oriented human-com- puter interaction would include the development of systems for maximizing the capture, retrieval, and display of clinically rel- evant information and handling related uncertainties in ways that minimally distract from attention to the patient and situation yet provide information in a manner that is immediately understand- able and interpretable. Such uncertainties include those associated with the information itself and those associated with other mat- ters as well, such as how a patient might respond to treatment or scientific uncertainties about the nature of a disease. Specialized systems would provide different presentations for the different relevant audiences: caregivers, medical staff, insurance compa- nies, patients, and relatives. The research challenge is to be able to extract information relevant to the moment in a way that can read- ily be assimilated from the tables, graphs, and free-text informa- tion about the patient. The collection and recording of information should be incorporated into the normal examination and caregiv- ing actions so that these actions do not disrupt caregiving (as is the case now), yet provide a comprehensive record. Information dashboards would allow a rapid overview of multiple patients, calling attention to cases that require closer examination. As before, the committee did not address in this short report the nature or scale of support needed and believes that this is an issue best addressed in a second phase of this study. • Encourage (or at least do not impede) efforts by health care organizations and communities to aggregate data about health care people, processes, and out- comes from all sources subject to appropriate protection of priacy and confiden- tiality. Data aggregation efforts, which should be regarded as infrastruc- tural in nature, will entail some expense, and reimbursement schedules should not discourage such expenses. Recognize that the time for payoff from these systems may be lengthy, while a critical mass of data is being acquired, while data quality is improved, and while systems and pro- cesses are developed that can utilize the data. Encourage the decoupling of data from applications (e.g., more reimbursement might be allowed for organizations that have the capability to export data in standard formats that accommodate heterogeneous data types). Where possible, reduce or eliminate organizational and legal barriers to data sharing while taking due note of relevant privacy concerns.

OCR for page 59
 RECOMMENDATIONS • Support additional education and training efforts at the intersection of health care, computer science, and health/biomedical informatics. The purpose of such efforts is to produce more individuals with expertise in both domains—physicians or nurses with undergraduate or graduate degrees in computer science or industrial and systems engineering, computer science researchers knowledgeable about medicine (e.g., with a master’s degree in medical innovation) who work on health care problems, and so on. The National Institutes of Health career development programs (often known as the K program) and institutional training programs for medi- cal informatics are models for such support,4 as are the research training programs in health/biomedical informatics supported by the National Library of Medicine at many educational organizations in the United States.5 6.2 THE COMPUTER SCIENCE COMMUNITY As early as 1992, the computer science community was exhorted to seek intellectual challenges in problem domains of societal significance. 6 Nowhere are such challenges more apparent and important than in health care. Accordingly, the committee believes that the computer science com- munity should: • Engage as co-equal intellectual partners and collaborators with health care practitioners and experts in health/biomedical informatics and other releant disciplines, such as industrial and process engineering and design, in an ongoing relationship to understand and sole problems of importance to health care. Such engagement will require overcoming important differences of intellectual style that inevitably separate disciplines. For example, there may be intel- lectual tensions between simplification and abstraction in the service of understanding on the one hand and the capture of details in the service of clinical fidelity on the other—and such tensions will have both positive and negative consequences. • Deelop institutional mechanisms within academia for rewarding work at the health care/computer science interface. As argued in other reports,7 institutional difficulties often arise in academia when interdisciplinary 4See http://grants.nih.gov/training/careerdevelopmentawards.htm. 5See http://www.nlm.nih.gov/ep/GrantTrainInstitute.html. 6National Research Council, Computing the Future, National Academy Press, Washington, D.C., 1992, available online at http://www.nap.edu/catalog.php?record_id=1982. 7See, for example, National Research Council, Catalyzing Inquiry at the Interface of Comput- ing and Biology, 2005; or National Research Council, Fostering Research on the Economic and Social Impacts of Information Technology, National Academy Press, Washington, D.C., 1998.

OCR for page 59
 COMPUTATIONAL TECHNOLOGY FOR EFFECTIVE HEALTH CARE work is involved. Collaborators from different disciplines must find and maintain common ground, such as agreeing on goals for a joint project, but must also respect one another’s separate priorities, such as having to publish in primary journals, present at particular conferences, or obtain tenure in their respective departments according to departmental crite- ria. Such cross-pressures and expectations from home departments and disciplinary colleagues remain even if the participants in a collaboration have similar goals for a project. (An example might be the Harvard-MIT program in health sciences and technology.) • Support educational and retraining efforts for computer science research- ers who want to explore research opportunities in health care. Such efforts might be offered across a broad front and might span a range in several dimensions, including time and format (e.g., weeks to years; courses, workshops, degree programs, postdoctoral fellowships), content (i.e., dif- ferent problems within health care), and target audience (i.e., undergradu- ates to fully tenured faculty). 6.3 HEALTH CARE ORGANIzATIONS The senior management in health care organizations (including the chief executive officer, chief quality officer, chief medical informatics offi- cer, chief information officer, and chief financial officer) and health care payers have often taken the lead in the deployment of IT for health care and are thus the primary audience to whom the following recommenda- tions are directed. • Organize incenties, roles, workflow, processes, and supporting infra- structure to encourage, support, and respond to opportunities for clinical perfor- mance gains. Focus on identifying, prioritizing, and managing changes in process and workflow, and only after doing so support them by technol- ogy. Use the context of the organization’s quality improvement strategy to guide and correct IT decisions. • Balance the institution’s IT portfolio among the four domains of automa- tion, connectiity, decision support, and data-mining capabilities. • Deelop the necessary data infrastructure for health care improement by aggregating data regarding people, processes, and outcomes from all sources. • Insist that endors supply IT that permits the separation of data from applications and facilitates data transfers to and from other non-endor applica- tions in shareable and generally useful formats. • Seek IT solutions that yield incremental gains from incremental efforts. Efforts that make progress in many small steps build support and con- sensus from the grass roots. One example of such an approach might

OCR for page 59
 RECOMMENDATIONS be an institutional commitment to digitize all paper records and make them available electronically in image format to all care providers. Even if capturing paper records in such a form would not make all of their content machine readable, it would go a long way toward eliminating the widely acknowledged problem of record unavailability that plagues a large number of patient-provider visits. And the infrastructure needed for such efforts could be used in the future to support other applications.